4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * #!-checking implemented by tytso.
11 * Demand-loading implemented 01.12.91 - no need to read anything but
12 * the header into memory. The inode of the executable is put into
13 * "current->executable", and page faults do the actual loading. Clean.
15 * Once more I can proudly say that linux stood up to being changed: it
16 * was less than 2 hours work to get demand-loading completely implemented.
18 * Demand loading changed July 1993 by Eric Youngdale. Use mmap instead,
19 * current->executable is only used by the procfs. This allows a dispatch
20 * table to check for several different types of binary formats. We keep
21 * trying until we recognize the file or we run out of supported binary
25 #include <linux/slab.h>
26 #include <linux/file.h>
27 #include <linux/fdtable.h>
29 #include <linux/stat.h>
30 #include <linux/fcntl.h>
31 #include <linux/smp_lock.h>
32 #include <linux/swap.h>
33 #include <linux/string.h>
34 #include <linux/init.h>
35 #include <linux/pagemap.h>
36 #include <linux/highmem.h>
37 #include <linux/spinlock.h>
38 #include <linux/key.h>
39 #include <linux/personality.h>
40 #include <linux/binfmts.h>
41 #include <linux/utsname.h>
42 #include <linux/pid_namespace.h>
43 #include <linux/module.h>
44 #include <linux/namei.h>
45 #include <linux/proc_fs.h>
46 #include <linux/mount.h>
47 #include <linux/security.h>
48 #include <linux/syscalls.h>
49 #include <linux/tsacct_kern.h>
50 #include <linux/cn_proc.h>
51 #include <linux/audit.h>
52 #include <linux/tracehook.h>
53 #include <linux/kmod.h>
55 #include <asm/uaccess.h>
56 #include <asm/mmu_context.h>
60 /* for /sbin/loader handling in search_binary_handler() */
61 #include <linux/a.out.h>
65 char core_pattern[CORENAME_MAX_SIZE] = "core";
66 int suid_dumpable = 0;
68 /* The maximal length of core_pattern is also specified in sysctl.c */
70 static LIST_HEAD(formats);
71 static DEFINE_RWLOCK(binfmt_lock);
73 int register_binfmt(struct linux_binfmt * fmt)
77 write_lock(&binfmt_lock);
78 list_add(&fmt->lh, &formats);
79 write_unlock(&binfmt_lock);
83 EXPORT_SYMBOL(register_binfmt);
85 void unregister_binfmt(struct linux_binfmt * fmt)
87 write_lock(&binfmt_lock);
89 write_unlock(&binfmt_lock);
92 EXPORT_SYMBOL(unregister_binfmt);
94 static inline void put_binfmt(struct linux_binfmt * fmt)
96 module_put(fmt->module);
100 * Note that a shared library must be both readable and executable due to
103 * Also note that we take the address to load from from the file itself.
105 asmlinkage long sys_uselib(const char __user * library)
109 char *tmp = getname(library);
110 int error = PTR_ERR(tmp);
113 error = path_lookup_open(AT_FDCWD, tmp,
115 FMODE_READ|FMODE_EXEC);
122 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
126 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
129 error = vfs_permission(&nd, MAY_READ | MAY_EXEC | MAY_OPEN);
133 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
134 error = PTR_ERR(file);
140 struct linux_binfmt * fmt;
142 read_lock(&binfmt_lock);
143 list_for_each_entry(fmt, &formats, lh) {
144 if (!fmt->load_shlib)
146 if (!try_module_get(fmt->module))
148 read_unlock(&binfmt_lock);
149 error = fmt->load_shlib(file);
150 read_lock(&binfmt_lock);
152 if (error != -ENOEXEC)
155 read_unlock(&binfmt_lock);
161 release_open_intent(&nd);
168 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
174 #ifdef CONFIG_STACK_GROWSUP
176 ret = expand_stack_downwards(bprm->vma, pos);
181 ret = get_user_pages(current, bprm->mm, pos,
182 1, write, 1, &page, NULL);
187 unsigned long size = bprm->vma->vm_end - bprm->vma->vm_start;
191 * We've historically supported up to 32 pages (ARG_MAX)
192 * of argument strings even with small stacks
198 * Limit to 1/4-th the stack size for the argv+env strings.
200 * - the remaining binfmt code will not run out of stack space,
201 * - the program will have a reasonable amount of stack left
204 rlim = current->signal->rlim;
205 if (size > rlim[RLIMIT_STACK].rlim_cur / 4) {
214 static void put_arg_page(struct page *page)
219 static void free_arg_page(struct linux_binprm *bprm, int i)
223 static void free_arg_pages(struct linux_binprm *bprm)
227 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
230 flush_cache_page(bprm->vma, pos, page_to_pfn(page));
233 static int __bprm_mm_init(struct linux_binprm *bprm)
236 struct vm_area_struct *vma = NULL;
237 struct mm_struct *mm = bprm->mm;
239 bprm->vma = vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
243 down_write(&mm->mmap_sem);
247 * Place the stack at the largest stack address the architecture
248 * supports. Later, we'll move this to an appropriate place. We don't
249 * use STACK_TOP because that can depend on attributes which aren't
252 vma->vm_end = STACK_TOP_MAX;
253 vma->vm_start = vma->vm_end - PAGE_SIZE;
255 vma->vm_flags = VM_STACK_FLAGS;
256 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
257 err = insert_vm_struct(mm, vma);
259 up_write(&mm->mmap_sem);
263 mm->stack_vm = mm->total_vm = 1;
264 up_write(&mm->mmap_sem);
266 bprm->p = vma->vm_end - sizeof(void *);
273 kmem_cache_free(vm_area_cachep, vma);
279 static bool valid_arg_len(struct linux_binprm *bprm, long len)
281 return len <= MAX_ARG_STRLEN;
286 static struct page *get_arg_page(struct linux_binprm *bprm, unsigned long pos,
291 page = bprm->page[pos / PAGE_SIZE];
292 if (!page && write) {
293 page = alloc_page(GFP_HIGHUSER|__GFP_ZERO);
296 bprm->page[pos / PAGE_SIZE] = page;
302 static void put_arg_page(struct page *page)
306 static void free_arg_page(struct linux_binprm *bprm, int i)
309 __free_page(bprm->page[i]);
310 bprm->page[i] = NULL;
314 static void free_arg_pages(struct linux_binprm *bprm)
318 for (i = 0; i < MAX_ARG_PAGES; i++)
319 free_arg_page(bprm, i);
322 static void flush_arg_page(struct linux_binprm *bprm, unsigned long pos,
327 static int __bprm_mm_init(struct linux_binprm *bprm)
329 bprm->p = PAGE_SIZE * MAX_ARG_PAGES - sizeof(void *);
333 static bool valid_arg_len(struct linux_binprm *bprm, long len)
335 return len <= bprm->p;
338 #endif /* CONFIG_MMU */
341 * Create a new mm_struct and populate it with a temporary stack
342 * vm_area_struct. We don't have enough context at this point to set the stack
343 * flags, permissions, and offset, so we use temporary values. We'll update
344 * them later in setup_arg_pages().
346 int bprm_mm_init(struct linux_binprm *bprm)
349 struct mm_struct *mm = NULL;
351 bprm->mm = mm = mm_alloc();
356 err = init_new_context(current, mm);
360 err = __bprm_mm_init(bprm);
376 * count() counts the number of strings in array ARGV.
378 static int count(char __user * __user * argv, int max)
386 if (get_user(p, argv))
400 * 'copy_strings()' copies argument/environment strings from the old
401 * processes's memory to the new process's stack. The call to get_user_pages()
402 * ensures the destination page is created and not swapped out.
404 static int copy_strings(int argc, char __user * __user * argv,
405 struct linux_binprm *bprm)
407 struct page *kmapped_page = NULL;
409 unsigned long kpos = 0;
417 if (get_user(str, argv+argc) ||
418 !(len = strnlen_user(str, MAX_ARG_STRLEN))) {
423 if (!valid_arg_len(bprm, len)) {
428 /* We're going to work our way backwords. */
434 int offset, bytes_to_copy;
436 offset = pos % PAGE_SIZE;
440 bytes_to_copy = offset;
441 if (bytes_to_copy > len)
444 offset -= bytes_to_copy;
445 pos -= bytes_to_copy;
446 str -= bytes_to_copy;
447 len -= bytes_to_copy;
449 if (!kmapped_page || kpos != (pos & PAGE_MASK)) {
452 page = get_arg_page(bprm, pos, 1);
459 flush_kernel_dcache_page(kmapped_page);
460 kunmap(kmapped_page);
461 put_arg_page(kmapped_page);
464 kaddr = kmap(kmapped_page);
465 kpos = pos & PAGE_MASK;
466 flush_arg_page(bprm, kpos, kmapped_page);
468 if (copy_from_user(kaddr+offset, str, bytes_to_copy)) {
477 flush_kernel_dcache_page(kmapped_page);
478 kunmap(kmapped_page);
479 put_arg_page(kmapped_page);
485 * Like copy_strings, but get argv and its values from kernel memory.
487 int copy_strings_kernel(int argc,char ** argv, struct linux_binprm *bprm)
490 mm_segment_t oldfs = get_fs();
492 r = copy_strings(argc, (char __user * __user *)argv, bprm);
496 EXPORT_SYMBOL(copy_strings_kernel);
501 * During bprm_mm_init(), we create a temporary stack at STACK_TOP_MAX. Once
502 * the binfmt code determines where the new stack should reside, we shift it to
503 * its final location. The process proceeds as follows:
505 * 1) Use shift to calculate the new vma endpoints.
506 * 2) Extend vma to cover both the old and new ranges. This ensures the
507 * arguments passed to subsequent functions are consistent.
508 * 3) Move vma's page tables to the new range.
509 * 4) Free up any cleared pgd range.
510 * 5) Shrink the vma to cover only the new range.
512 static int shift_arg_pages(struct vm_area_struct *vma, unsigned long shift)
514 struct mm_struct *mm = vma->vm_mm;
515 unsigned long old_start = vma->vm_start;
516 unsigned long old_end = vma->vm_end;
517 unsigned long length = old_end - old_start;
518 unsigned long new_start = old_start - shift;
519 unsigned long new_end = old_end - shift;
520 struct mmu_gather *tlb;
522 BUG_ON(new_start > new_end);
525 * ensure there are no vmas between where we want to go
528 if (vma != find_vma(mm, new_start))
532 * cover the whole range: [new_start, old_end)
534 vma_adjust(vma, new_start, old_end, vma->vm_pgoff, NULL);
537 * move the page tables downwards, on failure we rely on
538 * process cleanup to remove whatever mess we made.
540 if (length != move_page_tables(vma, old_start,
541 vma, new_start, length))
545 tlb = tlb_gather_mmu(mm, 0);
546 if (new_end > old_start) {
548 * when the old and new regions overlap clear from new_end.
550 free_pgd_range(tlb, new_end, old_end, new_end,
551 vma->vm_next ? vma->vm_next->vm_start : 0);
554 * otherwise, clean from old_start; this is done to not touch
555 * the address space in [new_end, old_start) some architectures
556 * have constraints on va-space that make this illegal (IA64) -
557 * for the others its just a little faster.
559 free_pgd_range(tlb, old_start, old_end, new_end,
560 vma->vm_next ? vma->vm_next->vm_start : 0);
562 tlb_finish_mmu(tlb, new_end, old_end);
565 * shrink the vma to just the new range.
567 vma_adjust(vma, new_start, new_end, vma->vm_pgoff, NULL);
572 #define EXTRA_STACK_VM_PAGES 20 /* random */
575 * Finalizes the stack vm_area_struct. The flags and permissions are updated,
576 * the stack is optionally relocated, and some extra space is added.
578 int setup_arg_pages(struct linux_binprm *bprm,
579 unsigned long stack_top,
580 int executable_stack)
583 unsigned long stack_shift;
584 struct mm_struct *mm = current->mm;
585 struct vm_area_struct *vma = bprm->vma;
586 struct vm_area_struct *prev = NULL;
587 unsigned long vm_flags;
588 unsigned long stack_base;
590 #ifdef CONFIG_STACK_GROWSUP
591 /* Limit stack size to 1GB */
592 stack_base = current->signal->rlim[RLIMIT_STACK].rlim_max;
593 if (stack_base > (1 << 30))
594 stack_base = 1 << 30;
596 /* Make sure we didn't let the argument array grow too large. */
597 if (vma->vm_end - vma->vm_start > stack_base)
600 stack_base = PAGE_ALIGN(stack_top - stack_base);
602 stack_shift = vma->vm_start - stack_base;
603 mm->arg_start = bprm->p - stack_shift;
604 bprm->p = vma->vm_end - stack_shift;
606 stack_top = arch_align_stack(stack_top);
607 stack_top = PAGE_ALIGN(stack_top);
608 stack_shift = vma->vm_end - stack_top;
610 bprm->p -= stack_shift;
611 mm->arg_start = bprm->p;
615 bprm->loader -= stack_shift;
616 bprm->exec -= stack_shift;
618 down_write(&mm->mmap_sem);
619 vm_flags = VM_STACK_FLAGS;
622 * Adjust stack execute permissions; explicitly enable for
623 * EXSTACK_ENABLE_X, disable for EXSTACK_DISABLE_X and leave alone
624 * (arch default) otherwise.
626 if (unlikely(executable_stack == EXSTACK_ENABLE_X))
628 else if (executable_stack == EXSTACK_DISABLE_X)
629 vm_flags &= ~VM_EXEC;
630 vm_flags |= mm->def_flags;
632 ret = mprotect_fixup(vma, &prev, vma->vm_start, vma->vm_end,
638 /* Move stack pages down in memory. */
640 ret = shift_arg_pages(vma, stack_shift);
642 up_write(&mm->mmap_sem);
647 #ifdef CONFIG_STACK_GROWSUP
648 stack_base = vma->vm_end + EXTRA_STACK_VM_PAGES * PAGE_SIZE;
650 stack_base = vma->vm_start - EXTRA_STACK_VM_PAGES * PAGE_SIZE;
652 ret = expand_stack(vma, stack_base);
657 up_write(&mm->mmap_sem);
660 EXPORT_SYMBOL(setup_arg_pages);
662 #endif /* CONFIG_MMU */
664 struct file *open_exec(const char *name)
670 err = path_lookup_open(AT_FDCWD, name, LOOKUP_FOLLOW, &nd,
671 FMODE_READ|FMODE_EXEC);
676 if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
679 if (nd.path.mnt->mnt_flags & MNT_NOEXEC)
682 err = vfs_permission(&nd, MAY_EXEC | MAY_OPEN);
686 file = nameidata_to_filp(&nd, O_RDONLY|O_LARGEFILE);
690 err = deny_write_access(file);
699 release_open_intent(&nd);
704 EXPORT_SYMBOL(open_exec);
706 int kernel_read(struct file *file, unsigned long offset,
707 char *addr, unsigned long count)
715 /* The cast to a user pointer is valid due to the set_fs() */
716 result = vfs_read(file, (void __user *)addr, count, &pos);
721 EXPORT_SYMBOL(kernel_read);
723 static int exec_mmap(struct mm_struct *mm)
725 struct task_struct *tsk;
726 struct mm_struct * old_mm, *active_mm;
728 /* Notify parent that we're no longer interested in the old VM */
730 old_mm = current->mm;
731 mm_release(tsk, old_mm);
735 * Make sure that if there is a core dump in progress
736 * for the old mm, we get out and die instead of going
737 * through with the exec. We must hold mmap_sem around
738 * checking core_state and changing tsk->mm.
740 down_read(&old_mm->mmap_sem);
741 if (unlikely(old_mm->core_state)) {
742 up_read(&old_mm->mmap_sem);
747 active_mm = tsk->active_mm;
750 activate_mm(active_mm, mm);
752 arch_pick_mmap_layout(mm);
754 up_read(&old_mm->mmap_sem);
755 BUG_ON(active_mm != old_mm);
756 mm_update_next_owner(old_mm);
765 * This function makes sure the current process has its own signal table,
766 * so that flush_signal_handlers can later reset the handlers without
767 * disturbing other processes. (Other processes might share the signal
768 * table via the CLONE_SIGHAND option to clone().)
770 static int de_thread(struct task_struct *tsk)
772 struct signal_struct *sig = tsk->signal;
773 struct sighand_struct *oldsighand = tsk->sighand;
774 spinlock_t *lock = &oldsighand->siglock;
777 if (thread_group_empty(tsk))
778 goto no_thread_group;
781 * Kill all other threads in the thread group.
784 if (signal_group_exit(sig)) {
786 * Another group action in progress, just
787 * return so that the signal is processed.
789 spin_unlock_irq(lock);
792 sig->group_exit_task = tsk;
793 zap_other_threads(tsk);
795 /* Account for the thread group leader hanging around: */
796 count = thread_group_leader(tsk) ? 1 : 2;
797 sig->notify_count = count;
798 while (atomic_read(&sig->count) > count) {
799 __set_current_state(TASK_UNINTERRUPTIBLE);
800 spin_unlock_irq(lock);
804 spin_unlock_irq(lock);
807 * At this point all other threads have exited, all we have to
808 * do is to wait for the thread group leader to become inactive,
809 * and to assume its PID:
811 if (!thread_group_leader(tsk)) {
812 struct task_struct *leader = tsk->group_leader;
814 sig->notify_count = -1; /* for exit_notify() */
816 write_lock_irq(&tasklist_lock);
817 if (likely(leader->exit_state))
819 __set_current_state(TASK_UNINTERRUPTIBLE);
820 write_unlock_irq(&tasklist_lock);
825 * The only record we have of the real-time age of a
826 * process, regardless of execs it's done, is start_time.
827 * All the past CPU time is accumulated in signal_struct
828 * from sister threads now dead. But in this non-leader
829 * exec, nothing survives from the original leader thread,
830 * whose birth marks the true age of this process now.
831 * When we take on its identity by switching to its PID, we
832 * also take its birthdate (always earlier than our own).
834 tsk->start_time = leader->start_time;
836 BUG_ON(!same_thread_group(leader, tsk));
837 BUG_ON(has_group_leader_pid(tsk));
839 * An exec() starts a new thread group with the
840 * TGID of the previous thread group. Rehash the
841 * two threads with a switched PID, and release
842 * the former thread group leader:
845 /* Become a process group leader with the old leader's pid.
846 * The old leader becomes a thread of the this thread group.
847 * Note: The old leader also uses this pid until release_task
848 * is called. Odd but simple and correct.
850 detach_pid(tsk, PIDTYPE_PID);
851 tsk->pid = leader->pid;
852 attach_pid(tsk, PIDTYPE_PID, task_pid(leader));
853 transfer_pid(leader, tsk, PIDTYPE_PGID);
854 transfer_pid(leader, tsk, PIDTYPE_SID);
855 list_replace_rcu(&leader->tasks, &tsk->tasks);
857 tsk->group_leader = tsk;
858 leader->group_leader = tsk;
860 tsk->exit_signal = SIGCHLD;
862 BUG_ON(leader->exit_state != EXIT_ZOMBIE);
863 leader->exit_state = EXIT_DEAD;
864 write_unlock_irq(&tasklist_lock);
866 release_task(leader);
869 sig->group_exit_task = NULL;
870 sig->notify_count = 0;
874 flush_itimer_signals();
876 if (atomic_read(&oldsighand->count) != 1) {
877 struct sighand_struct *newsighand;
879 * This ->sighand is shared with the CLONE_SIGHAND
880 * but not CLONE_THREAD task, switch to the new one.
882 newsighand = kmem_cache_alloc(sighand_cachep, GFP_KERNEL);
886 atomic_set(&newsighand->count, 1);
887 memcpy(newsighand->action, oldsighand->action,
888 sizeof(newsighand->action));
890 write_lock_irq(&tasklist_lock);
891 spin_lock(&oldsighand->siglock);
892 rcu_assign_pointer(tsk->sighand, newsighand);
893 spin_unlock(&oldsighand->siglock);
894 write_unlock_irq(&tasklist_lock);
896 __cleanup_sighand(oldsighand);
899 BUG_ON(!thread_group_leader(tsk));
904 * These functions flushes out all traces of the currently running executable
905 * so that a new one can be started
907 static void flush_old_files(struct files_struct * files)
912 spin_lock(&files->file_lock);
914 unsigned long set, i;
918 fdt = files_fdtable(files);
919 if (i >= fdt->max_fds)
921 set = fdt->close_on_exec->fds_bits[j];
924 fdt->close_on_exec->fds_bits[j] = 0;
925 spin_unlock(&files->file_lock);
926 for ( ; set ; i++,set >>= 1) {
931 spin_lock(&files->file_lock);
934 spin_unlock(&files->file_lock);
937 char *get_task_comm(char *buf, struct task_struct *tsk)
939 /* buf must be at least sizeof(tsk->comm) in size */
941 strncpy(buf, tsk->comm, sizeof(tsk->comm));
946 void set_task_comm(struct task_struct *tsk, char *buf)
949 strlcpy(tsk->comm, buf, sizeof(tsk->comm));
953 int flush_old_exec(struct linux_binprm * bprm)
957 char tcomm[sizeof(current->comm)];
960 * Make sure we have a private signal table and that
961 * we are unassociated from the previous thread group.
963 retval = de_thread(current);
967 set_mm_exe_file(bprm->mm, bprm->file);
970 * Release all of the old mmap stuff
972 retval = exec_mmap(bprm->mm);
976 bprm->mm = NULL; /* We're using it now */
978 /* This is the point of no return */
979 current->sas_ss_sp = current->sas_ss_size = 0;
981 if (current->euid == current->uid && current->egid == current->gid)
982 set_dumpable(current->mm, 1);
984 set_dumpable(current->mm, suid_dumpable);
986 name = bprm->filename;
988 /* Copies the binary name from after last slash */
989 for (i=0; (ch = *(name++)) != '\0';) {
991 i = 0; /* overwrite what we wrote */
993 if (i < (sizeof(tcomm) - 1))
997 set_task_comm(current, tcomm);
999 current->flags &= ~PF_RANDOMIZE;
1002 /* Set the new mm task size. We have to do that late because it may
1003 * depend on TIF_32BIT which is only updated in flush_thread() on
1004 * some architectures like powerpc
1006 current->mm->task_size = TASK_SIZE;
1008 if (bprm->e_uid != current->euid || bprm->e_gid != current->egid) {
1010 set_dumpable(current->mm, suid_dumpable);
1011 current->pdeath_signal = 0;
1012 } else if (file_permission(bprm->file, MAY_READ) ||
1013 (bprm->interp_flags & BINPRM_FLAGS_ENFORCE_NONDUMP)) {
1015 set_dumpable(current->mm, suid_dumpable);
1018 /* An exec changes our domain. We are no longer part of the thread
1021 current->self_exec_id++;
1023 flush_signal_handlers(current, 0);
1024 flush_old_files(current->files);
1032 EXPORT_SYMBOL(flush_old_exec);
1035 * Fill the binprm structure from the inode.
1036 * Check permissions, then read the first 128 (BINPRM_BUF_SIZE) bytes
1038 int prepare_binprm(struct linux_binprm *bprm)
1041 struct inode * inode = bprm->file->f_path.dentry->d_inode;
1044 mode = inode->i_mode;
1045 if (bprm->file->f_op == NULL)
1048 bprm->e_uid = current->euid;
1049 bprm->e_gid = current->egid;
1051 if(!(bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)) {
1053 if (mode & S_ISUID) {
1054 current->personality &= ~PER_CLEAR_ON_SETID;
1055 bprm->e_uid = inode->i_uid;
1060 * If setgid is set but no group execute bit then this
1061 * is a candidate for mandatory locking, not a setgid
1064 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) {
1065 current->personality &= ~PER_CLEAR_ON_SETID;
1066 bprm->e_gid = inode->i_gid;
1070 /* fill in binprm security blob */
1071 retval = security_bprm_set(bprm);
1075 memset(bprm->buf,0,BINPRM_BUF_SIZE);
1076 return kernel_read(bprm->file,0,bprm->buf,BINPRM_BUF_SIZE);
1079 EXPORT_SYMBOL(prepare_binprm);
1081 static int unsafe_exec(struct task_struct *p)
1083 int unsafe = tracehook_unsafe_exec(p);
1085 if (atomic_read(&p->fs->count) > 1 ||
1086 atomic_read(&p->files->count) > 1 ||
1087 atomic_read(&p->sighand->count) > 1)
1088 unsafe |= LSM_UNSAFE_SHARE;
1093 void compute_creds(struct linux_binprm *bprm)
1097 if (bprm->e_uid != current->uid) {
1099 current->pdeath_signal = 0;
1104 unsafe = unsafe_exec(current);
1105 security_bprm_apply_creds(bprm, unsafe);
1106 task_unlock(current);
1107 security_bprm_post_apply_creds(bprm);
1109 EXPORT_SYMBOL(compute_creds);
1112 * Arguments are '\0' separated strings found at the location bprm->p
1113 * points to; chop off the first by relocating brpm->p to right after
1114 * the first '\0' encountered.
1116 int remove_arg_zero(struct linux_binprm *bprm)
1119 unsigned long offset;
1127 offset = bprm->p & ~PAGE_MASK;
1128 page = get_arg_page(bprm, bprm->p, 0);
1133 kaddr = kmap_atomic(page, KM_USER0);
1135 for (; offset < PAGE_SIZE && kaddr[offset];
1136 offset++, bprm->p++)
1139 kunmap_atomic(kaddr, KM_USER0);
1142 if (offset == PAGE_SIZE)
1143 free_arg_page(bprm, (bprm->p >> PAGE_SHIFT) - 1);
1144 } while (offset == PAGE_SIZE);
1153 EXPORT_SYMBOL(remove_arg_zero);
1156 * cycle the list of binary formats handler, until one recognizes the image
1158 int search_binary_handler(struct linux_binprm *bprm,struct pt_regs *regs)
1160 unsigned int depth = bprm->recursion_depth;
1162 struct linux_binfmt *fmt;
1164 /* handle /sbin/loader.. */
1166 struct exec * eh = (struct exec *) bprm->buf;
1168 if (!bprm->loader && eh->fh.f_magic == 0x183 &&
1169 (eh->fh.f_flags & 0x3000) == 0x3000)
1172 unsigned long loader;
1174 allow_write_access(bprm->file);
1178 loader = bprm->vma->vm_end - sizeof(void *);
1180 file = open_exec("/sbin/loader");
1181 retval = PTR_ERR(file);
1185 /* Remember if the application is TASO. */
1186 bprm->taso = eh->ah.entry < 0x100000000UL;
1189 bprm->loader = loader;
1190 retval = prepare_binprm(bprm);
1193 /* should call search_binary_handler recursively here,
1194 but it does not matter */
1198 retval = security_bprm_check(bprm);
1202 /* kernel module loader fixup */
1203 /* so we don't try to load run modprobe in kernel space. */
1206 retval = audit_bprm(bprm);
1211 for (try=0; try<2; try++) {
1212 read_lock(&binfmt_lock);
1213 list_for_each_entry(fmt, &formats, lh) {
1214 int (*fn)(struct linux_binprm *, struct pt_regs *) = fmt->load_binary;
1217 if (!try_module_get(fmt->module))
1219 read_unlock(&binfmt_lock);
1220 retval = fn(bprm, regs);
1222 * Restore the depth counter to its starting value
1223 * in this call, so we don't have to rely on every
1224 * load_binary function to restore it on return.
1226 bprm->recursion_depth = depth;
1229 tracehook_report_exec(fmt, bprm, regs);
1231 allow_write_access(bprm->file);
1235 current->did_exec = 1;
1236 proc_exec_connector(current);
1239 read_lock(&binfmt_lock);
1241 if (retval != -ENOEXEC || bprm->mm == NULL)
1244 read_unlock(&binfmt_lock);
1248 read_unlock(&binfmt_lock);
1249 if (retval != -ENOEXEC || bprm->mm == NULL) {
1251 #ifdef CONFIG_MODULES
1253 #define printable(c) (((c)=='\t') || ((c)=='\n') || (0x20<=(c) && (c)<=0x7e))
1254 if (printable(bprm->buf[0]) &&
1255 printable(bprm->buf[1]) &&
1256 printable(bprm->buf[2]) &&
1257 printable(bprm->buf[3]))
1258 break; /* -ENOEXEC */
1259 request_module("binfmt-%04x", *(unsigned short *)(&bprm->buf[2]));
1266 EXPORT_SYMBOL(search_binary_handler);
1268 void free_bprm(struct linux_binprm *bprm)
1270 free_arg_pages(bprm);
1275 * sys_execve() executes a new program.
1277 int do_execve(char * filename,
1278 char __user *__user *argv,
1279 char __user *__user *envp,
1280 struct pt_regs * regs)
1282 struct linux_binprm *bprm;
1284 struct files_struct *displaced;
1287 retval = unshare_files(&displaced);
1292 bprm = kzalloc(sizeof(*bprm), GFP_KERNEL);
1296 file = open_exec(filename);
1297 retval = PTR_ERR(file);
1304 bprm->filename = filename;
1305 bprm->interp = filename;
1307 retval = bprm_mm_init(bprm);
1311 bprm->argc = count(argv, MAX_ARG_STRINGS);
1312 if ((retval = bprm->argc) < 0)
1315 bprm->envc = count(envp, MAX_ARG_STRINGS);
1316 if ((retval = bprm->envc) < 0)
1319 retval = security_bprm_alloc(bprm);
1323 retval = prepare_binprm(bprm);
1327 retval = copy_strings_kernel(1, &bprm->filename, bprm);
1331 bprm->exec = bprm->p;
1332 retval = copy_strings(bprm->envc, envp, bprm);
1336 retval = copy_strings(bprm->argc, argv, bprm);
1340 current->flags &= ~PF_KTHREAD;
1341 retval = search_binary_handler(bprm,regs);
1343 /* execve success */
1344 security_bprm_free(bprm);
1345 acct_update_integrals(current);
1348 put_files_struct(displaced);
1354 security_bprm_free(bprm);
1362 allow_write_access(bprm->file);
1370 reset_files_struct(displaced);
1375 int set_binfmt(struct linux_binfmt *new)
1377 struct linux_binfmt *old = current->binfmt;
1380 if (!try_module_get(new->module))
1383 current->binfmt = new;
1385 module_put(old->module);
1389 EXPORT_SYMBOL(set_binfmt);
1391 /* format_corename will inspect the pattern parameter, and output a
1392 * name into corename, which must have space for at least
1393 * CORENAME_MAX_SIZE bytes plus one byte for the zero terminator.
1395 static int format_corename(char *corename, long signr)
1397 const char *pat_ptr = core_pattern;
1398 int ispipe = (*pat_ptr == '|');
1399 char *out_ptr = corename;
1400 char *const out_end = corename + CORENAME_MAX_SIZE;
1402 int pid_in_pattern = 0;
1404 /* Repeat as long as we have more pattern to process and more output
1407 if (*pat_ptr != '%') {
1408 if (out_ptr == out_end)
1410 *out_ptr++ = *pat_ptr++;
1412 switch (*++pat_ptr) {
1415 /* Double percent, output one percent */
1417 if (out_ptr == out_end)
1424 rc = snprintf(out_ptr, out_end - out_ptr,
1425 "%d", task_tgid_vnr(current));
1426 if (rc > out_end - out_ptr)
1432 rc = snprintf(out_ptr, out_end - out_ptr,
1433 "%d", current->uid);
1434 if (rc > out_end - out_ptr)
1440 rc = snprintf(out_ptr, out_end - out_ptr,
1441 "%d", current->gid);
1442 if (rc > out_end - out_ptr)
1446 /* signal that caused the coredump */
1448 rc = snprintf(out_ptr, out_end - out_ptr,
1450 if (rc > out_end - out_ptr)
1454 /* UNIX time of coredump */
1457 do_gettimeofday(&tv);
1458 rc = snprintf(out_ptr, out_end - out_ptr,
1460 if (rc > out_end - out_ptr)
1467 down_read(&uts_sem);
1468 rc = snprintf(out_ptr, out_end - out_ptr,
1469 "%s", utsname()->nodename);
1471 if (rc > out_end - out_ptr)
1477 rc = snprintf(out_ptr, out_end - out_ptr,
1478 "%s", current->comm);
1479 if (rc > out_end - out_ptr)
1483 /* core limit size */
1485 rc = snprintf(out_ptr, out_end - out_ptr,
1486 "%lu", current->signal->rlim[RLIMIT_CORE].rlim_cur);
1487 if (rc > out_end - out_ptr)
1497 /* Backward compatibility with core_uses_pid:
1499 * If core_pattern does not include a %p (as is the default)
1500 * and core_uses_pid is set, then .%pid will be appended to
1501 * the filename. Do not do this for piped commands. */
1502 if (!ispipe && !pid_in_pattern && core_uses_pid) {
1503 rc = snprintf(out_ptr, out_end - out_ptr,
1504 ".%d", task_tgid_vnr(current));
1505 if (rc > out_end - out_ptr)
1514 static int zap_process(struct task_struct *start)
1516 struct task_struct *t;
1519 start->signal->flags = SIGNAL_GROUP_EXIT;
1520 start->signal->group_stop_count = 0;
1524 if (t != current && t->mm) {
1525 sigaddset(&t->pending.signal, SIGKILL);
1526 signal_wake_up(t, 1);
1529 } while_each_thread(start, t);
1534 static inline int zap_threads(struct task_struct *tsk, struct mm_struct *mm,
1535 struct core_state *core_state, int exit_code)
1537 struct task_struct *g, *p;
1538 unsigned long flags;
1541 spin_lock_irq(&tsk->sighand->siglock);
1542 if (!signal_group_exit(tsk->signal)) {
1543 mm->core_state = core_state;
1544 tsk->signal->group_exit_code = exit_code;
1545 nr = zap_process(tsk);
1547 spin_unlock_irq(&tsk->sighand->siglock);
1548 if (unlikely(nr < 0))
1551 if (atomic_read(&mm->mm_users) == nr + 1)
1554 * We should find and kill all tasks which use this mm, and we should
1555 * count them correctly into ->nr_threads. We don't take tasklist
1556 * lock, but this is safe wrt:
1559 * None of sub-threads can fork after zap_process(leader). All
1560 * processes which were created before this point should be
1561 * visible to zap_threads() because copy_process() adds the new
1562 * process to the tail of init_task.tasks list, and lock/unlock
1563 * of ->siglock provides a memory barrier.
1566 * The caller holds mm->mmap_sem. This means that the task which
1567 * uses this mm can't pass exit_mm(), so it can't exit or clear
1571 * It does list_replace_rcu(&leader->tasks, ¤t->tasks),
1572 * we must see either old or new leader, this does not matter.
1573 * However, it can change p->sighand, so lock_task_sighand(p)
1574 * must be used. Since p->mm != NULL and we hold ->mmap_sem
1577 * Note also that "g" can be the old leader with ->mm == NULL
1578 * and already unhashed and thus removed from ->thread_group.
1579 * This is OK, __unhash_process()->list_del_rcu() does not
1580 * clear the ->next pointer, we will find the new leader via
1584 for_each_process(g) {
1585 if (g == tsk->group_leader)
1587 if (g->flags & PF_KTHREAD)
1592 if (unlikely(p->mm == mm)) {
1593 lock_task_sighand(p, &flags);
1594 nr += zap_process(p);
1595 unlock_task_sighand(p, &flags);
1599 } while_each_thread(g, p);
1603 atomic_set(&core_state->nr_threads, nr);
1607 static int coredump_wait(int exit_code, struct core_state *core_state)
1609 struct task_struct *tsk = current;
1610 struct mm_struct *mm = tsk->mm;
1611 struct completion *vfork_done;
1614 init_completion(&core_state->startup);
1615 core_state->dumper.task = tsk;
1616 core_state->dumper.next = NULL;
1617 core_waiters = zap_threads(tsk, mm, core_state, exit_code);
1618 up_write(&mm->mmap_sem);
1620 if (unlikely(core_waiters < 0))
1624 * Make sure nobody is waiting for us to release the VM,
1625 * otherwise we can deadlock when we wait on each other
1627 vfork_done = tsk->vfork_done;
1629 tsk->vfork_done = NULL;
1630 complete(vfork_done);
1634 wait_for_completion(&core_state->startup);
1636 return core_waiters;
1639 static void coredump_finish(struct mm_struct *mm)
1641 struct core_thread *curr, *next;
1642 struct task_struct *task;
1644 next = mm->core_state->dumper.next;
1645 while ((curr = next) != NULL) {
1649 * see exit_mm(), curr->task must not see
1650 * ->task == NULL before we read ->next.
1654 wake_up_process(task);
1657 mm->core_state = NULL;
1661 * set_dumpable converts traditional three-value dumpable to two flags and
1662 * stores them into mm->flags. It modifies lower two bits of mm->flags, but
1663 * these bits are not changed atomically. So get_dumpable can observe the
1664 * intermediate state. To avoid doing unexpected behavior, get get_dumpable
1665 * return either old dumpable or new one by paying attention to the order of
1666 * modifying the bits.
1668 * dumpable | mm->flags (binary)
1669 * old new | initial interim final
1670 * ---------+-----------------------
1678 * (*) get_dumpable regards interim value of 10 as 11.
1680 void set_dumpable(struct mm_struct *mm, int value)
1684 clear_bit(MMF_DUMPABLE, &mm->flags);
1686 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1689 set_bit(MMF_DUMPABLE, &mm->flags);
1691 clear_bit(MMF_DUMP_SECURELY, &mm->flags);
1694 set_bit(MMF_DUMP_SECURELY, &mm->flags);
1696 set_bit(MMF_DUMPABLE, &mm->flags);
1701 int get_dumpable(struct mm_struct *mm)
1705 ret = mm->flags & 0x3;
1706 return (ret >= 2) ? 2 : ret;
1709 int do_coredump(long signr, int exit_code, struct pt_regs * regs)
1711 struct core_state core_state;
1712 char corename[CORENAME_MAX_SIZE + 1];
1713 struct mm_struct *mm = current->mm;
1714 struct linux_binfmt * binfmt;
1715 struct inode * inode;
1718 int fsuid = current->fsuid;
1721 unsigned long core_limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
1722 char **helper_argv = NULL;
1723 int helper_argc = 0;
1726 audit_core_dumps(signr);
1728 binfmt = current->binfmt;
1729 if (!binfmt || !binfmt->core_dump)
1731 down_write(&mm->mmap_sem);
1733 * If another thread got here first, or we are not dumpable, bail out.
1735 if (mm->core_state || !get_dumpable(mm)) {
1736 up_write(&mm->mmap_sem);
1741 * We cannot trust fsuid as being the "true" uid of the
1742 * process nor do we know its entire history. We only know it
1743 * was tainted so we dump it as root in mode 2.
1745 if (get_dumpable(mm) == 2) { /* Setuid core dump mode */
1746 flag = O_EXCL; /* Stop rewrite attacks */
1747 current->fsuid = 0; /* Dump root private */
1750 retval = coredump_wait(exit_code, &core_state);
1755 * Clear any false indication of pending signals that might
1756 * be seen by the filesystem code called to write the core file.
1758 clear_thread_flag(TIF_SIGPENDING);
1761 * lock_kernel() because format_corename() is controlled by sysctl, which
1762 * uses lock_kernel()
1765 ispipe = format_corename(corename, signr);
1768 * Don't bother to check the RLIMIT_CORE value if core_pattern points
1769 * to a pipe. Since we're not writing directly to the filesystem
1770 * RLIMIT_CORE doesn't really apply, as no actual core file will be
1771 * created unless the pipe reader choses to write out the core file
1772 * at which point file size limits and permissions will be imposed
1773 * as it does with any other process
1775 if ((!ispipe) && (core_limit < binfmt->min_coredump))
1779 helper_argv = argv_split(GFP_KERNEL, corename+1, &helper_argc);
1780 /* Terminate the string before the first option */
1781 delimit = strchr(corename, ' ');
1784 delimit = strrchr(helper_argv[0], '/');
1788 delimit = helper_argv[0];
1789 if (!strcmp(delimit, current->comm)) {
1790 printk(KERN_NOTICE "Recursive core dump detected, "
1795 core_limit = RLIM_INFINITY;
1797 /* SIGPIPE can happen, but it's just never processed */
1798 if (call_usermodehelper_pipe(corename+1, helper_argv, NULL,
1800 printk(KERN_INFO "Core dump to %s pipe failed\n",
1805 file = filp_open(corename,
1806 O_CREAT | 2 | O_NOFOLLOW | O_LARGEFILE | flag,
1810 inode = file->f_path.dentry->d_inode;
1811 if (inode->i_nlink > 1)
1812 goto close_fail; /* multiple links - don't dump */
1813 if (!ispipe && d_unhashed(file->f_path.dentry))
1816 /* AK: actually i see no reason to not allow this for named pipes etc.,
1817 but keep the previous behaviour for now. */
1818 if (!ispipe && !S_ISREG(inode->i_mode))
1821 * Dont allow local users get cute and trick others to coredump
1822 * into their pre-created files:
1824 if (inode->i_uid != current->fsuid)
1828 if (!file->f_op->write)
1830 if (!ispipe && do_truncate(file->f_path.dentry, 0, 0, file) != 0)
1833 retval = binfmt->core_dump(signr, regs, file, core_limit);
1836 current->signal->group_exit_code |= 0x80;
1838 filp_close(file, NULL);
1841 argv_free(helper_argv);
1843 current->fsuid = fsuid;
1844 coredump_finish(mm);